To satisfy the recommended slew rate of 5ns/V (which is 25ns/5V), the resistor would have to be 6.2k, which will mean a considerable current draw when the comparator is sinking.
Why is this a concern?
a) if I can prevent wasting yet another milliamp (here and there) in the stable state, I prefer to do it, within reason, of course. Current limitations aren't too low, this will be powered from USB, so kinda sorta 500 mA if we play by the rules of the non-PD ports, even though I know that most USB PSUs will deliver 1 to 2 amps no questions asked. Of this 500 mA, however, most will be dedicated to a power-hungry subcircuit, so the lower the consumption of the logic part will be, the better.
b) as I said, it may turn out that the resistor's value will be irrelevant. I will have to measure the actual LM393's transition time with different resistors, and I expect it to be in or close to the microsecond range -- well outside the non-Schmitt 74LVC inputs recommended range. I'll post some oscillograms of that here for the sake of completeness when I do it.
I would normally use LM393 with 1-6mA I_OL to best take advantage of its speed (the edge rate is ca. 200ns; the delay is only ~µs for very small (~mV) input changes -- or, it's a knock-off "clone" that's different altogether; I have seen CMOS parts masquerading under deceptive part numbers before, e.g. 3Peak Inc. LM2903A-SR*, beware!). You didn't mention any supply current limitations so I'm going to assume it's a non-issue. Using something as hungry as LM393 also seems suspect to that end (but, it is one of the most available chips worldwide, and perhaps better alternatives aren't).
Yeah, well, again, will need to test my specific parts, which are supposed to be genuine TI ones. I've never tested how fast they switch, and now I have a good excuse for that.
For 5V comparator duty, I normally reach for MCP6562, a nice low power, faster, CMOS output device. I've seen it go in and out of stock over the years, and maybe it's not very available in your region, I don't know.
Yes, that's a good idea, and for this specific task a CMOS output comparator (rather than an open collector one) will actually be a better choice.
Thanks for the tip on the specific IC, which, as it turns out, is actually sold here, and in the convenient SOIC-8 package too. Not super cheap at just below $1 apiece, but I think I will still grab some to use where CMOS output can be better than open collector.
I also have some TLV3501s, but I don't want to waste them on this particular device, these are going to wait for something that actually requires high speed.
May also be an option to (ab)use LVDS receivers as comparators -- if they have a wide input CM range (and most are rail to rail), that should be well enough behaved, and as I understand it, they regularly outperform the rather pessimistic 100mV spec that LVDS requires -- particularly if you can [self-, even?]calibrate out the offset in the application, or a little offset is acceptable.
Good to keep in mind as a possible technique (yet another!), but in this case nah, it's simply not worth it, when compared to using two additional buffers (or the SC70 74LVC1G132 gate).